Electrical switching of an antiferromagnet

See allHide authors and affiliations

Science  05 Feb 2016:
Vol. 351, Issue 6273, pp. 587-590
DOI: 10.1126/science.aab1031

Manipulating a stubborn magnet

Spintronics is an alternative to conventional electronics, based on using the electron's spin rather than its charge. Spintronic devices, such as magnetic memory, have traditionally used ferromagnetic materials to encode the 1's and 0's of the binary code. A weakness of this approach—that strong magnetic fields can erase the encoded information—could be avoided by using antiferromagnets instead of ferromagnets. But manipulating the magnetic ordering of antiferromagnets is tricky. Now, Wadley et al. have found a way (see the Perspective by Marrows). Running currents along specific directions in the thin films of the antiferromagnetic compound CuMnAs reoriented the magnetic domains in the material.

Science, this issue p. 587; see also p. 558


Antiferromagnets are hard to control by external magnetic fields because of the alternating directions of magnetic moments on individual atoms and the resulting zero net magnetization. However, relativistic quantum mechanics allows for generating current-induced internal fields whose sign alternates with the periodicity of the antiferromagnetic lattice. Using these fields, which couple strongly to the antiferromagnetic order, we demonstrate room-temperature electrical switching between stable configurations in antiferromagnetic CuMnAs thin-film devices by applied current with magnitudes of order 106 ampere per square centimeter. Electrical writing is combined in our solid-state memory with electrical readout and the stored magnetic state is insensitive to and produces no external magnetic field perturbations, which illustrates the unique merits of antiferromagnets for spintronics.

In charge-based information devices, perturbations such as ionizing radiation can lead to data loss. In contrast, spin-based devices, in which different magnetic moment orientations in a ferromagnet (FM) represent the zeros and ones (1), are robust against charge perturbations. However, the FM moments can be unintentionally reoriented and the data erased by perturbing magnetic fields generated externally or internally within the memory circuitry. If magnetic memories were based on antiferromagnets (AFMs) instead, they would be robust against charge and magnetic field perturbations. Additional advantages of AFMs compared to FMs include the invisibility of data stored in AFMs to external magnetic probes, ultrafast spin dynamics in AFMs, and the broad range of metal, semiconductor, or insulator materials with room-temperature AFM order (27).

The energy barrier separating stable orientations of ordered spins is due to the magnetic anisotropy energy. It is an even function of the magnetic moment, which implies that the magnetic anisotropy and the corresponding memory functionality are readily present in both FMs and AFMs (8, 9). The magneto-transport counterpart of the magnetic anisotropy energy is the anisotropic magnetoresistance (AMR). In the early 1990s, the first generation of FM magnetic random access memory (MRAM) microdevices used AMR for the electrical readout of the memory state (10). AMR is an even function of the magnetic moment, which again implies its presence in AFMs (11). Although AMR in AFMs was experimentally confirmed in several recent studies (1217), efficient means for manipulating AFM moments have remained elusive.

It has been proposed that current-induced spin transfer torques of the form Embedded Image, which are used for electrical writing in the most advanced FM MRAMs (1), could also produce large-angle reorientation of the AFM moments (18). In these antidamping-like torques, Embedded Image is the magnetic moment vector and Embedded Image is the electrically injected carrier spin polarization. Translated to AFMs, the effective field proportional to Embedded Image that drives the antidamping-like torque Embedded Image on individual spin sublattices A and B has the favorable staggered property, i.e., alternates in sign between the opposite spin sublattices.

In FM spin-transfer-torque MRAMs, spin-polarized carriers are injected into the free FM layer from a fixed FM polarizer by an out-of-plane electrical current driven through the FM-FM stack. In analogy, (18) assumes injection of the spin-polarized carriers into the AFM from a fixed FM polarizer by out-of-plane electrical current driven in a FM-AFM stack. However, relativistic spin-orbit coupling may offer staggered current-induced fields, which do not require external polarizers and which act in bare AFM crystals (19). The effect occurs in AFMs with specific crystal and magnetic structures for which the spin sublattices form space-inversion partners. Among these materials is a high–Néel temperature AFM, tetragonal-phase CuMnAs, which was recently synthesized in the form of single-crystal epilayers on III-V semiconductor substrates (20).

Relativistic current-induced fields observed previously in broken inversion-symmetry FM crystals (2129) can originate from the inverse spin-galvanic effect (3034) (Fig. 1, A and B). The full lattice of the CuMnAs crystal (Fig. 1C) has an inversion symmetry with the center of inversion at an interstitial position (green ball in the figure). This implies that the mechanism described in Fig. 1, A and B, will not generate a net current-induced spin density when integrated over the entire crystal. However, Mn atoms form two sublattices (depicted in Fig. 1C in red and purple) whose local environment has broken inversion symmetry, and the two Mn sublattices form inversion partners. The inverse spin-galvanic mechanisms of Fig. 1, A and B, will generate locally nonequilibrium spin polarizations of opposite signs on the inversion-partner Mn sublattices. For these staggered fields to couple strongly to the AFM order, it is essential that the inversion-partner Mn sublattices coincide with the two spin sublattices A and B of the AFM ground state (19). The resulting spin-orbit torques have the form Embedded Image, where the effective field proportional to Embedded Image acting on the spin-sublattice magnetizations Embedded Imagealternates in sign between the two sublattices. The CuMnAs crystal and magnetic structures (Fig. 1C) fulfill these symmetry requirements (20).

Fig. 1 Theory of the staggered current-induced field in CuMnAs.

(A) Schematic of the inverse spin-galvanic effect in a model inversion asymmetric Rashba spin texture (red arrows). Embedded Image are the in-plane momentum components. The nonequilibrium redistribution of carriers from the left side to the right side of the Fermi surface results in a net in-plane spin polarization (thick red arrow) along Embedded Image direction, where Embedded Image is the applied current (black arrow). (B) Same as (A) for opposite sense of the inversion asymmetry, resulting in a net in-plane spin polarization (thick purple arrow) along Embedded Image direction. (C) CuMnAs crystal structure and AFM ordering. The two Mn spin-sublattices A and B (red and purple) are inversion partners. This and panels A and B imply opposite sign of the respective local current–induced spin polarizations, Embedded Image, at spin sublattices A and B. The full CuMnAs crystal is centrosymmetric around the interstitial position highlighted by the green ball. (D) Microscopic calculations of the components of the spin-orbit field transverse to the magnetic moments per current density 107 A cm−2 at spin sublattices A and B as a function of the magnetic moment angle Embedded Imageφ measured from the x axis ([100] crystal direction). The electrical current is applied along the x and y axes.

To quantitatively estimate the strength of the staggered current-induced field, we performed microscopic calculations based on the Kubo linear response formalism (35) (see supplementary text for details). The calculations (Fig. 1D) confirm the desired opposite sign of the current-induced field on the two spin sublattices and highlight the expected dependence on the magnetic moment angle, which implies that the AFM moments will tend to align perpendicular to the applied current. For reversible electrical switching between two stable states and the subsequent electrical detection by the AMR, the setting current pulses can therefore be applied along two orthogonal in-plane cubic axes of CuMnAs. The magnitude of the effect seen in Fig. 1D is comparable to that of typical current-induced fields applied in FMs, suggesting that CuMnAs is a favorable material for observing current-induced switching in an AFM.

Our experiments were conducted on epitaxial films of the tetragonal phase of CuMnAs, which is a member of a broad family of high-temperature I-Mn-V AFM compounds (6, 7, 20). We have observed the electrical switching and readout effects described below in more than 20 devices fabricated from five different CuMnAs films, with thicknesses ranging from 40 to 80 nm, grown on either GaP or GaAs substrates. The electrical data shown in Figs. 2 to 4 were obtained on a 46-nm epilayer on lattice-matched GaP(001), whose transmission electron microscopy image (Fig. 2A) demonstrates excellent structural and chemical order (20). Consistent with the AFM order of the CuMnAs film, superconducting quantum interference device (SQUID) magnetometry measurements (Fig. 2B) show only the diamagnetic background of the sample substrate. X-ray magnetic linear dichroism–photoelectron emission microscopy (XMLD-PEEM) measurements at the Mn L3 absorption edge show that the AFM moments are oriented in the plane of the film, with a submicrometer-scale domain structure (Fig. 2C). Neutron diffraction confirmed collinear AFM order with a Néel temperature Embedded Image K (20, 36). The CuMnAs film is metallic with a room-temperature sheet resistivity of 160 microhm cm.

Fig. 2 Electrical switching of the AFM CuMnAs.

(A) Scanning transmission electron microscopy image of CuMnAs/GaP in the [100]–[001] plane. (B) Magnetization versus applied field of an unpatterned piece of the CuMnAs/GaP wafer measured by SQUID magnetometer. (C) XMLD-PEEM image of the CuMnAs film with x-rays at the Mn L3 absorption edge incident at 16° from the surface along the [100] axis. (D) Optical microscopy image of the device and schematic of the measurement geometry. (E) Change in the transverse resistance after applying three successive 50-ms writing pulses of amplitude Embedded Image A cmEmbedded Image alternately along the [100] crystal direction of CuMnAs (black arrow in panel D and black points in panel E) and along the [010] axis (red arrow in panel D and red points in panel E). The reading current Embedded Image is applied along the [Embedded Image10] axis, and transverse resistance signals Embedded Image are recorded 10 s after each writing pulse. A constant offset is subtracted from Embedded Image. Measurements were done at a sample temperature of 273 K.

Fig. 3 Dependence of the switching on the writing pulse length and amplitude.

Transverse resistance after successive writing pulses along the [100] axis (black points) and [010] axis (red points) for different current amplitudes (A) or pulse lengths (B). Embedded Image is recorded 10 s after each writing pulse. Embedded Image is the average of the longitudinal resistance Embedded Image. Measurements were done at sample temperature of 273 K. A constant offset is subtracted from Embedded Image.

In Fig. 2E, we demonstrate the electrical writing in a CuMnAs device (Fig. 2D). The sample was held at a stable temperature of 273 K inferred from the temperature calibration of the resistivity of the CuMnAs film. Three successive 50-ms writing pulses of amplitude Embedded Image AcmEmbedded Image were applied alternately along the [100] crystal axis of CuMnAs (black arrow in Fig. 2D and black points in Fig. 2E) and along the [010] axis (red arrow in Fig. 2D and red points in Fig. 2E). Note that Embedded Image A cmEmbedded Image is the current density in the central region of the device obtained from finite element modeling for the applied current of 90 mA driven through the 28-Embedded Imagem-wide writing arms of the device. The reading current Embedded Image was applied along the [Embedded Image] in-plane diagonal and resistance signals, Embedded Image, transverse to Embedded Image are recorded 10 s after each Embedded Image pulse. A constant offset is subtracted from Embedded Image.

The [100]-directed writing pulses are expected to set a preference for domains with AFM spin axis along the [010] direction (black double-arrow in Fig. 2D) and the [010]-directed pulses for domains with AFM spin axis along the [100] direction (red double-arrow in Fig. 2D). Consistent with this picture, successive Embedded Image pulses in one direction increase the amplitude of the readout Embedded Image signal of one sign and pulsing in the orthogonal direction increases the amplitude of Embedded Image of the opposite sign. As seen in Fig. 2E, all the AFM memory states can be written reproducibly. The signals are independent of the polarity of the writing current, which is expected for the current-induced switching in AFMs. The amplitude of the switching current applied in our AFM memory is comparable to that of FM spin transfer torque MRAMs and is much lower than in the early observations of spin-orbit torque switching in FM metals, where 100 MA cmEmbedded Image pulses were used to reverse magnetization in a Pt/Co bilayer (29).

In Fig. 3, A and B, we explore in more detail the domain reconfiguration by applying a series of 50 Embedded Image pulses of varying length and amplitude along the [010] direction (red points) and [100] direction (black points) at 273 K. The data, which again show highly reproducible switching patterns, illustrate that the imbalance in the domain populations increases with the length and amplitude of the writing pulses and tends to saturate with the increasing number of pulses. Because in these measurements, heating of the central region of the device can reach tens of degrees during the writing pulses, we did not explore the switching behavior further beyond the pulse lengths and amplitudes shown in Fig. 3, A and B. More intense pulses in our device design can lead to irreproducible characteristics or device failure due to structural changes. Apart from the absolute Embedded Image values, we also indicate in Fig. 3, A and B, relative values Embedded Image of the signal, where Embedded Image is the longitudinal resistance Embedded Image averaged over the different states set by the writing pulses along the [100]/[010] directions. Below we will associate Embedded Image, reaching 0.2%, with the transverse AFM AMR. Further confirmation of the picture of the current-induced domain reconfiguration by the applied writing pulses is given by XMLD-PEEM measurements and XMLD spectroscopy (see figs. S1 and S2 and supplementary text.)

Fig. 4 AMR symmetry of the electrical readout signals.

(A) Optical microscopy image of the device and of the measurement geometries with different probe current directions (green arrows). The writing current directions are shown by black and red arrows. (B) Normalized transverse resistance Embedded Image after five writing current pulses along the [100] axis (black) and five pulses along the [010] axis (red) for the reading current directions shown in (A). Vertical lines indicate the times of the pulses. The pulse length is 275 ms and amplitude Embedded Image A cmEmbedded Image. Measurements were done at a sample temperature of 150 K. A constant offset is subtracted from Embedded Image. (C) As for (B) but for the normalized longitudinal resistance change, Embedded Image, where Embedded Image.

We now analyze the symmetry of the measured resistances for different probe current directions. Figure 4 shows switching data for both the transverse resistance signal and the longitudinal signal, Embedded Image, where Embedded Image, obtained at the sample temperature of 150 K. In these lower-temperature experiments, we applied five successive 275-ms pulses of amplitude Embedded Image A cmEmbedded Image along the [100] or [010] axis to obtain signals comparable to the higher-temperature measurements. Each row in Fig. 4 corresponds to a different axis along which we apply the probe current Embedded Image. From top to bottom, the reading current is applied along the crystal axis [1Embedded Image0], [110], [100], and [010].

Consistent with the AMR symmetry, the transverse signals (also known as the planar Hall effect) are detected for the AFM spin-axes angle set toward Embedded Image from the probe current, and the transverse signal flips sign when the probe current is rotated by Embedded Image. The corresponding longitudinal signals vanish in this geometry. For AFM spin axes set toward Embedded Image from the probe current, the transverse signal vanishes and the longitudinal signal is detected, which is again consistent with the AMR symmetries. The AMR nature of the electrical signals is further confirmed by the comparable amplitudes of the transverse and longitudinal signals. We note that apart from the stable AMR signals, the longitudinal resistances show an additional time dependence, which is due to the cooling of the sample after the writing pulses. These isotropic changes in Embedded Image correspond to a temperature change of a fraction of a Kelvin over the probing time interval.

We also observe these AMR symmetries in higher-temperature measurements. However, the AMR changes sign between the higher- and lower-temperature data, as seen when comparing the transverse resistance signals in Figs. 2 and 3 with the corresponding measurements in the first row of Fig. 4. The change in sign of the AMR is further confirmed in fig. S3 (see also the supplementary text), where the measured temperature dependence of AMR is shown and compared to calculations. From this comparison, we can infer the preferred AFM spin-axis direction for the given writing current direction. The experimental and theoretical AMR signs match if the AFM spin axis aligns perpendicular to the writing current. This is consistent with the predicted direction of the spin-orbit current-induced fields and with the XMLD-PEEM results. Measurements at high magnetic fields shown in fig. S3 (see also supplementary text) give further confirmation that the AFM spin axis aligns perpendicular to the setting current pulses. These measurements also highlight that our AFM memory can be read and written by the staggered current-induced fields and the memory state retained even in the presence of strong magnetic fields.

The staggered current-induced fields that we observe are not unique to CuMnAs. The high–Néel temperature AFM Mn2Au (37) is another example in which the spin sublattices form inversion partners and where theory predicts large field-like torques of the form Embedded Image with Embedded Image (19). From our microscopic density-functional calculations, we obtain a current-induced field of around 20 Oe per 10Embedded Image A cmEmbedded Image in Mn2Au, which, combined with its higher conductivity, may make this a favorable system for observing current-driven AFM switching. AFMs that do not possess these specific symmetries can in principle be switched by injecting a spin current into the AFM from a spin-orbit–coupled nonmagnetic (NM) layer using an applied in-plane electrical current via the spin Hall effect, generating the antidamping-like torque Embedded Image (19). The same type of torque can be generated by the spin-orbit Berry-curvature mechanism acting at the inversion-asymmetric AFM/NM interface or in bare AFM crystals with globally noncentrosymmetric unit cells like CuMnSb (19). Our experiments in CuMnAs, combined with the prospect of other realizations of these relativistic nonequilibrium phenomena in AFMs, indicate that AFMs are now ready to join the rapidly developing fields of basic and applied spintronics, enriching this area of solid-state physics and microelectronics by the range of unique characteristics of AFMs.

Supplementary Materials

Supplementary Text

Figs. S1 to S3

References (38, 39)

References and Notes

  1. Acknowledgments: We acknowledge support from the European Union (EU) European Research Council Advanced (grant 268066); the Ministry of Education of the Czech Republic (grant LM2011026); the Grant Agency of the Czech Republic (grant 14-37427); the UK Engineering and Physical Sciences Research Council (grant EP/K027808/1); the EU 7th Framework Programme (grant REGPOT-CT-2013-316014 and FP7-People-2012-ITN-316657); HGF Programme VH-NG 513 and Deutsche Forschungsgemeinschaft SPP 1568; supercomputing resources at Jülich Supercomputing Centre and RWTH Aachen University; and Diamond Light Source for the allocation of beamtime under proposal number SI-12504. We thank C. Nelson for providing the scanning transmission electron microscopy measurement.
View Abstract

Navigate This Article